Date Approved
1-12-2026
Embargo Period
1-12-2028
Document Type
Dissertation
Degree Name
Ph.D. Data Science
Department
Computer Science
College
College of Science & Mathematics
Advisor
Bo Sun, Ph.D.
Committee Member 1
Anthony Breitzman, Ph.D.
Committee Member 2
Shen-Shyang Ho, Ph.D.
Committee Member 3
Guimu Guo, Ph.D.
Committee Member 4
Ning Wang, Ph.D.
Keywords
Explainable Artificial Intelligence;Reinforcement Learning;Visual Analytics
Disciplines
Computer Sciences | Physical Sciences and Mathematics
Abstract
Understanding how Reinforcement Learning (RL) agents make decisions remains a critical challenge in advancing transparent and trustworthy artificial intelligence (AI). Explainable AI (XAI) methods have been introduced to open the black-box models. However, XAI in RL is under-explored due to the challenges such as temporal dependencies, sequential decision-making, and dynamic policies. Moreover, most XAI methods focus on mathematical or feature-based explanations, which is non-intuitive and challenging for end-users who benefit from AI on a regular basis. This dissertation addresses these gaps by developing a visual-based exploration framework that extends traditional XAI toward interactive storytelling to enhance the interpretability and understanding of Deep RL (DRL). This approach used Bag of Pattern (BoP) to capture temporal policy summaries and recurring behavioral patterns over time along with their contributions to rewards. These extracted patterns provide valuable behavioral training insights, reveal how agent behavior evolves during learning, and suggest indicators for convergence and potential early stopping in DRL. The proposed tool supports both experts and end-users offering analytical insights and intuitive, human-centered visual explanations to enhance trustworthiness. Through case studies across environments with increasing complexity, the research demonstrates how visual storytelling can bridge the gap between agent decisions and human understanding.
Recommended Citation
Alicioglu, Gulsum, "A VISUAL EXPLORATION FRAMEWORK FOR EXPLAINABLE DEEP REINFORCEMENT LEARNING" (2026). Theses and Dissertations. 3473.
https://rdw.rowan.edu/etd/3473
